Cloudflare announced support for running Cursor Cloud Agents inside Cloudflare Sandboxes, adding another enterprise-controlled execution environment for AI coding agent workloads1,3.
The integration lets developers assign coding tasks from the Cursor app, cursor.com, or the Cursor mobile app, with tool calls — terminal commands, filesystem operations, and browser actions — executing inside sandbox environments within the customer's own Cloudflare account. Cursor continues to run the agent loop, including inference, planning, and orchestration, while the self-hosted worker handles execution on infrastructure the customer controls2.
Cloudflare's implementation places each assigned Cursor session in an isolated container. Repositories, build caches, and secrets remain on the customer's machines during self-hosted operation. File chunks read by the model during inference, along with artifacts such as screenshots, videos, and log references, are uploaded so they can appear in pull requests and dashboards.
The announcement arrives alongside a broader expansion of Cursor's self-hosted cloud agent infrastructure, which adds dynamically scheduled worker pools and support for running agent workloads across multiple cloud and infrastructure platforms. Cursor made self-hosted cloud agents generally available on March 25, 2026. The September update introduces team pools, where workers can serve requests from developers across a team or enterprise rather than being assigned to one individual machine or repository. Requests wait in a named pool until an available worker claims them, and separate pools can be configured for different execution environments, including GPU machines or Macs used for iOS development.
Cursor provides deployment options for Kubernetes and Google Cloud Run. Its Kubernetes operator manages warm capacity, rolling worker updates, and token rotation, while the Cloud Run deployment uses a custom autoscaler driven by the Cloud Agents API. Workers connect to Cursor through a long-lived outbound HTTPS connection; Cursor does not initiate inbound connections into the customer's network.
The update also adds computer-use support for self-hosted Linux workers alongside Mac machines. Once required desktop packages and Chrome or Chromium are installed, an agent can interact with a browser, take screenshots, and use other computer-control functions. Developers can view the agent's desktop while it is working or take control of the environment through Cursor.
Cursor says self-hosted workers can run across infrastructure from AWS Lambda, Cloudflare, Coder, Daytona, E2B, Modal, Namespace, and Vercel. Cursor-hosted virtual machines remain the default environment for its cloud agents.
Cloudflare described the integration as building on its work with other AI agent platforms, including Devin Outposts and Claude Managed Agents. Cloudflare has separately expanded its AI-adjacent infrastructure in recent weeks, deploying an adaptive bot detection engine that retrains its machine learning models continuously ctx and launching BotBase for Operators, a dashboard for bot directory submissions ctx.
ANALYSIS The Cloudflare Sandboxes integration and the broader self-hosted worker pool architecture address a consistent enterprise concern: keeping source code, credentials, and build systems off third-party infrastructure while still using cloud-hosted AI agent orchestration. The split — Cursor handles inference and planning, the customer handles execution — mirrors the pattern GitHub adopted when it added self-hosted runner support to Copilot coding agent in October 2025.
Teams can specify worker pools for Cloud Agents triggered through GitHub, Slack, and Linear, and automations configured through the Cloud Agents dashboard can target particular pools. Cursor also supports hibernating idle workers: organizations can snapshot and stop an idle machine, then restore the environment if an agent receives a follow-up request within a configured reconnect period.
Customers operating self-hosted machines remain responsible for worker images, infrastructure, secrets, scaling policies, and production validation.